Why NJi Started Working with vragen.ai

The reason was clear: AI is developing rapidly and directly impacts the role of a knowledge institute. When knowledge is your core task, you want to maintain control over how that knowledge is used and shared.

A demonstration of vragen.ai showed that this is possible. Thanks to a Retrieval-Augmented Generation (RAG) solution on its own content, NJi didn’t have to worry about 'black box' models or unclear data sources.

This gave confidence. Especially because vragen.ai demonstrably met the high requirements concerning security, privacy, and data control. NJi was thus able to safely experiment with AI without the risk of external models learning from their data.

NJi, along with Vragen.ai, developed their own AI search assistant: Ask NJi. A tool that answers questions based on its own knowledge, with source citations and within clear safety frameworks. And it helps visitors find their way through the wealth of information on the NJi website.

From No Early Adopter to AI Ambassador

Marleen van Riet, senior communications advisor at NJi, was initially hesitant about AI at the start of the project. She used it occasionally but didn’t feel like an ‘early adopter’. The technology didn’t immediately appeal, and the terms and workings were still unfamiliar. “How does generated AI work? What is a language model?”

In collaboration with the vragen.ai team, this changed. By helping build the tool, she gained insight into how to responsibly implement AI, how to maintain control over content, and how important that is within a knowledge organization like NJi.

"I now better understand what AI is and how to use it responsibly. And as a result, I am much more open to it. Including the ethical and societal issues."

- Marleen van Riet, senior communications advisor at NJi

The experience brought not only more understanding but also trust. Even an external AI event provided little new information: within NJi, the AI knowledge level has already risen significantly.

The Assistant as a Guide in an Overload of Information

Knowledge is NJi's core product. They want to continue defining, spreading, and managing it in a way that fits their content-related standards. With the continuous goal: to help every target group find the right answer faster and more effectively.

This underlying knowledge is carefully constructed, scientifically substantiated, and content-rich. That's precisely why reliability is one of the biggest focus points. If the AI assistant says something incorrect, it directly affects NJi's credibility. This was the biggest internal concern.

Therefore, a cautious approach was chosen. Not a free chatbot that 'engages in conversation', but not just a simple search engine either. Ask NJi is intentionally situated between these extremes. The assistant can provide short answers in some cases and mainly refers to reliable sources on the website. This way, the content remains within NJi's parameters.

Insights Per Target Group Based on Real Questions

NJi shares its knowledge in a target-group-oriented way: for professionals, policymakers, parents, youths, and education. Originally, the focus was mainly on professionals and policy. During the pandemic, it became clear that parents also had a strong need for reliable information. Since then, this target group has been structurally included. Youths are also increasingly given their own space in the offerings.

This impacts how knowledge is accessible. On the website, visitors now see content that aligns with their profile. This approach has been extended into the AI assistant. Each target group has a separate configuration: in tone, content, and relevancy of information. Users choose their perspective, and the assistant tailors the answers accordingly.

Simultaneously, NJi wants more insight: what are the target groups really looking for? To facilitate this, an additional data layer has been added. In the vragen.ai dashboard, it is now visible which questions come from which target group. Previously, this was not possible.

This makes it possible to monitor and analyze in a more targeted way. A separate cluster of employees is involved per target group, and the configuration is adjusted where necessary. This way, a clearer picture of what is happening and where knowledge is lacking is increasingly formed.

Designed for Safety and Reliability

The AI assistant is carefully designed with attention to safety, privacy, and reliability. For sensitive topics such as depression, suicide, or personal cases, the assistant follows extra guidelines. It does not go into content details but refers directly to appropriate help or the contact page. Personal data is automatically anonymized.

News reports are excluded because they become outdated quickly. And specialist databases such as guidelines, tools, and figures are findable by the AI but are not displayed in detail. Instead, the assistant refers to the source. This way, the context is preserved and risks surrounding interpretation are avoided.

Per topic and source, it is determined what role the AI may play: to show or not to show, to explain or not to explain, always with control.

What vragen.ai Reveals

"It is our core task to ensure that we have up-to-date knowledge. This is a flywheel that can stimulate this."

- Marleen van Riet, senior communications advisor at NJi

During the testing phase, it became clear that not all knowledge was equally current. Especially with figures, the AI sometimes relied on older sources because simply no newer data was available.

This led to a closer look at content topicality. The assistant immediately shows where information is outdated or lacking. For the communications team, this aids in signaling towards the organization. It helps to prioritize more effectively which files need renewal. And to improve content faster based on real questions from users.

From Scenarios to Go Live

Initially, the emphasis was on testing. Many ‘what-if’ scenarios were explored, with attention to risks and exceptions. What if someone enters a case? What if a sensitive subject is raised? What if data turns out to be outdated?

“We were stuck in the hypothetical scenarios. It became a quagmire of ‘what-if’ questions. That was the moment to collectively say it's time to go live.”

- Marleen van Riet, senior communications advisor at NJi

However, there comes a moment when you can't keep testing endlessly. Thanks to clear leadership from management, NJi dared to take that step: go live and learn from practice.

NJi now has an AI search assistant under its own control. It makes NJi's reliable knowledge more accessible and provides valuable insights per target group. At the same time, this accelerates the internal development of AI skills.

What NJi Wants to Convey to Other Organizations

See it as a strategic experiment, but one with direction. And think beforehand about monitoring: a live assistant also requires something from your organization.

“Ensure the right people from the organization are on board and participating in testing. For example, we really needed the content to be able to assess the answers. Then don't get bogged down in endless 'what if' testing. Ultimately, you get the best picture with real users of the questions actually being asked.

Set up monitoring properly, with clear processes, management, and ownership. This gives you insight into what is happening, how the tool works—and where your organization needs to take action.”

- Marleen van Riet, senior communications advisor at NJi

Looking Ahead: Step by Step Forward

In the coming time, NJi wants to build further. With better target group analyses based on practical questions. With structural improvements in the topicality of their content.

The assistant is now included in the website menu. It is still a pilot phase, in which NJi tests with real users how the tool functions and what is still needed. An evaluation will follow in January to determine how to continue its deployment.

In the future, the assistant could possibly replace the entire search function. That is not yet the case, but the foundation is there, and the process is iterative. Each step provides new insights.

Also Considering Developing Your Own AI Search Assistant?

Do you want to safely experiment with AI on your own knowledge, just like NJi, with full control over sources, answers, and target groups? Vragen.ai helps you with configuration, optimization, and monitoring.